Skip to content

Upgrading

Rolling upgrades from 8.x to 9.x

The Elasticsearch role supports zero-downtime rolling upgrades from Elastic 8.x to 9.x. Each node is stopped, upgraded, and restarted one at a time while the cluster stays operational.

How it works

  1. The role detects the installed version vs the target version
  2. For each node (one at a time):
    • Disables shard allocation to prevent rebalancing during restart
    • Flushes indices to reduce recovery time after restart
    • Stops Elasticsearch and upgrades the package
    • Starts Elasticsearch with the new version
    • Waits for the node to rejoin and the cluster to reach green
    • Re-enables shard allocation
  3. Kibana, Logstash, Beats, and Elastic Agent are upgraded after Elasticsearch

Running an upgrade

Change the release version and re-run your playbook:

group_vars/all.yml
elasticstack_release: 9   # was 8

ansible-playbook -i inventory.yml playbook.yml

The role handles the upgrade order automatically. Elasticsearch nodes are upgraded first, then Kibana, Logstash, Beats, and Elastic Agent.

Upgrade path requirement

Elasticsearch 9.x requires version 8.19.x as a stepping stone. The role validates this automatically — if you try to jump from 8.15 to 9.x, it fails with a clear error telling you to upgrade to 8.19 first. This is an Elastic requirement, not a role limitation.

Safety checks

The role refuses to upgrade if:

  • The cluster is not green before starting
  • The upgrade path requirement is not met (must be on 8.19.x for 9.x)

Unsafe restart mode

For development environments where you want to skip the safety checks:

elasticsearch_unsafe_upgrade_restart: true

Warning

Never use this in production. It bypasses the rolling upgrade safety checks and may cause data loss.

Upgrading other components

Kibana, Logstash, Beats, and Elastic Agent are simpler — they don't need rolling restarts. The roles detect the version mismatch, upgrade the package, and restart the service. Run them after all Elasticsearch nodes are upgraded. Package-based Elastic Agent upgrades are controlled by elasticstack_version; Fleet-managed binary upgrades require the Fleet-supported installation method.

Rollback

There is no automated rollback. If an upgrade fails:

  1. Stop Elasticsearch on the failed node
  2. Downgrade the package: apt install elasticsearch=8.x.y or yum downgrade elasticsearch-8.x.y
  3. Start Elasticsearch
  4. The node will rejoin the cluster with the old version

Note

Elasticsearch does not support running mixed major versions long-term. Complete the upgrade or roll back all nodes to the same version.

Minor version upgrades

Minor version upgrades (e.g. 8.15 to 8.17) are handled automatically. Re-run the playbook and the package manager installs the latest version within the major release. No special steps needed — the role handles service restart and health checks.